Andrej Karpathy joins Anthropic

Andrej Karpathy said on Tuesday, May 19, that he had joined Anthropic, a high-profile hire for the AI company behind Claude as competition for senior researchers intensifies. Karpathy is a founding member of OpenAI and a former senior director of AI at Tesla, making the move one of the most closely watched talent shifts in the sector this week. In a post on X, Karpathy wrote that “the next few years at the frontier of LLMs will be especially formative” and said he was “very excited” to return to research and development. Anthropic confirmed that he has started work on its pre-training effort, the part of model building that shapes a system’s core capabilities. ### Where exactly is Karpathy going inside Anthropic? Anthropic said Karpathy started this week on its pre-training team, working under team lead Nick Joseph. The company described pre-training as the large-scale training work that gives Claude its “core knowledge and capabilities,” one of the most compute-intensive stages in building a frontier model. (techcrunch.com) TechCrunch reported, citing an Anthropic spokesperson, that Karpathy will start a team focused on using Claude itself to speed up pre-training research. That would place him at a layer of development closer to foundational model creation than to product packaging or go-to-market work. ### Why did his post draw so much attention? (forbes.com) Karpathy has remained one of the best-known public figures in AI since leaving Tesla and, later, stepping back from OpenAI-related work. His X post was quickly picked up across tech and investor circles on May 19 and May 20, with outlets including CNBC, Axios and TechCrunch framing the move as a notable win for Anthropic in the contest for elite technical talent. (bestaitools.com) Karpathy wrote that he remains “deeply passionate about education” and plans to resume that work later, but said his immediate focus is R&D. That wording mattered because he had spent much of the recent period as a visible educator and commentator on AI, rather than in a full-time operating role at a frontier lab. (cnbc.com) ### Why is pre-training such a consequential assignment? Pre-training is the stage where companies run large, expensive training jobs to produce the base model later adapted for products and applications. Anthropic and other labs treat that layer as strategically important because it determines much of a model’s reasoning, coding and language performance before later tuning. (techcrunch.com) Karpathy’s background aligns with that work. CNBC and other outlets described him as an OpenAI co-founder who later led AI work at Tesla, experience that spans deep learning research and large-scale engineering. ### What does this say about Anthropic’s position right now? Anthropic has been expanding both model releases and infrastructure commitments in 2026. (msn.com) The company’s official site currently highlights Claude Opus 4.7, while its platform release notes say Claude Platform on AWS launched on May 11 with AWS billing and identity support. In April, Anthropic and Amazon said they were expanding their collaboration for up to 5 gigawatts of new data center capacity, and Anthropic said it would spend more than $100 billion on AWS technologies over 10 years. (cnbc.com) Anthropic said in its March 2025 funding announcement that it had raised $3.5 billion at a $61.5 billion post-money valuation. ### What happens next? (anthropic.com) Karpathy’s next visible step is likely to be inside Anthropic’s pre-training organization rather than at a public product launch. Anthropic has already tied that work to its Claude roadmap and to its AWS buildout, with significant Trainium2 capacity scheduled to come online in the second quarter and scaled Trainium3 capacity expected later this year, according to the company’s April announcement with Amazon. (anthropic.com)